Yiming Yang

Professor, Language Technologies Institute

Research

My research has centered on statistical learning methods/algorithms and application to very-large-scale text categorization, web-mining for concept graph discovery, semi-supervised clustering, multitask learning, novelty-based information retrieval, large-scale optimization for online advertising, social network analysis for personalized email prioritization, etc. My recent research focuses on the following topics:

Large-Scale Structured Learning for Hierarchical Classification  (Gopal & Yang, KDD 2013;  Gopal & Yang, ICML 2013 & Supplementary  ;  Gopal et al., NIPS 2012)

• Providing organizational views of multi-source Big Data (e.g., Wikipedia, online shops, Coursera)
• State-of-the-art classifiers for large-scale classification over hundreds of thousands of categories
• Scalable variational inference for joint optimization of one trillion (4 TB) model parameters

Scalable Machine Learning for Time Series Analysis (Topic Detection and Tracking)

• From scientific literature, news stories, sensor signals, maintenance reports, etc.
• Modeling multi-source and multi-scale evidence of dynamic chances in temporal sequences. (On-going NSF project; Gopal, PhD Thesis)
• A new family of Bayesian von Mices Fischer (vMF) clustering techniques (Gopal & Yang, ICML 2014 & Supplementary)
• Unsupervised clustering and semi-supervised metric learning and supervised classification (Gopal & Yang, UAI 2014 & Supplimentary).

Concept Graph Learning for Online Education (NSF project; Yang et al., WSDM 2015)

• Mapping online course materials to Wikipedia categories as the Interlingua (universal concepts)
• Predicting conceptual dependencies among courses based on partially observed prerequisites
• Planning customized curriculum for individuals based on backgrounds and goals

Macro-Level Information Fusion for Events and Entities (joint effort with Jaime Carbonell in the DARPA DEFT project)

• Detecting entities and events of interest in various forms of mentions in text to enable high-precision, semi-structured information fusion and summarization. Using a corporate acquisition event as an example, different (and partially redundant) sentences can mention acquirer, price, date, approvals, joint-management, etc. This multi-aspect information needs to be jointly extracted into a unified structured form for this event type, with uncertainty estimates in the aggregated representation.